# Agent Patterns Catalog

> A Pattern Language for Agentic Systems — a GoF-formal catalog of agentic patterns and agentic AI design patterns. 565 agent design patterns (also known as agentic design patterns, LLM agent design patterns, or agentic AI patterns) for building LLM agents across 14 books, with 253 mapped compositions (frameworks + recipes), 54 engineering methodologies, 90 training (enablement) patterns, and 10 curated framework pages. Each pattern is decomposed in the manner of Christopher Alexander (1977) and the Gang of Four (1994): context, problem, forces, therefore, solution, consequences, constrains, related. Licensed CC BY 4.0 — free to ingest, quote, and reason over with attribution.

Catalog version: unversioned. Last updated: see /patterns.json.

## Start here

- [Cover / contents](https://www.agentpatternscatalog.org/): The home page, with the language laid out as a book.
- [All patterns](https://www.agentpatternscatalog.org/patterns/): The full index of every pattern in the catalog, grouped by book.
- [Books](https://www.agentpatternscatalog.org/books/): Patterns grouped into Alexander-style books by theme.
- [Frameworks](https://www.agentpatternscatalog.org/frameworks/): Curated agent frameworks mapped to the patterns they natively support.
- [Compositions](https://www.agentpatternscatalog.org/compositions/): Upstream frameworks and recipes that map to the patterns they instantiate.
- [Methodologies](https://www.agentpatternscatalog.org/methodologies/): Step-by-step engineering methods (construction, evaluation, deployment, …) that compose the patterns into a build process.
- [Trainings](https://www.agentpatternscatalog.org/trainings/): Enablement patterns for upskilling a learner (human or autonomous agent) through experience, located along the Craft Path (foundation → operator → … → principal).
- [Compose](https://www.agentpatternscatalog.org/compose/): Assemble an agent from a curated recipe or framework starter.

## Wire it into your agent (MCP)

Live tools over a hosted Model Context Protocol (MCP) server. Point any MCP-capable client — Claude Code, Claude Desktop, Cursor, VS Code + Copilot, Codex, Cline, or any client that speaks MCP — at the endpoint below to call the catalog as live tools (search patterns, fetch a pattern by slug, list compositions, methodologies, trainings). Clients that don't speak MCP (ChatGPT, Claude.ai web, Gemini) can use the drop-in system prompt on /use/ instead.

- MCP endpoint: https://mcp.agentpatternscatalog.org/mcp
- Per-client setup (Claude Code, Cursor, VS Code, Codex, Cline, generic): https://www.agentpatternscatalog.org/use/

## Topic landings

Curated entry points, one per high-intent query. Each has its own h1, canonical URL, JSON-LD (CollectionPage + ItemList), and a distinct pattern selection — open the one that matches your framing.

- [Agentic Patterns](https://www.agentpatternscatalog.org/agentic-patterns/) — the pattern language for self-directed AI systems.
- [Agentic Design Patterns](https://www.agentpatternscatalog.org/agentic-design-patterns/) — autonomy / agency framing of the catalog.
- [Agentic AI Design Patterns](https://www.agentpatternscatalog.org/agentic-ai-design-patterns/) — production building blocks for agentic AI systems.
- [Agentic AI Architecture](https://www.agentpatternscatalog.org/agentic-ai-architecture/) — structural patterns for agentic AI systems.
- [AI Agent Design Patterns](https://www.agentpatternscatalog.org/ai-agent-design-patterns/) — how to build AI agents that reason, act, and stay safe.
- [LLM Agent Design Patterns](https://www.agentpatternscatalog.org/llm-agent-design-patterns/) — the LLM-era framing of the catalog.
- [Agent Design Patterns](https://www.agentpatternscatalog.org/agent-design-patterns/) — reusable shapes for software agents, LLM-era.
- [RAG Agent Patterns](https://www.agentpatternscatalog.org/rag-agent-patterns/) — retrieval-augmented patterns for agentic AI.
- [Multi-Agent Patterns](https://www.agentpatternscatalog.org/multi-agent-patterns/) — coordination patterns across multiple agents.
- [AI Agent Safety Patterns](https://www.agentpatternscatalog.org/ai-agent-safety-patterns/) — guardrail patterns for production agentic systems.

## Books in this language

- [anti-patterns](https://www.agentpatternscatalog.org/books/anti-patterns/): 117 patterns
- [cognition-introspection](https://www.agentpatternscatalog.org/books/cognition-introspection/): 26 patterns
- [governance-observability](https://www.agentpatternscatalog.org/books/governance-observability/): 47 patterns
- [memory](https://www.agentpatternscatalog.org/books/memory/): 35 patterns
- [multi-agent](https://www.agentpatternscatalog.org/books/multi-agent/): 50 patterns
- [planning-control-flow](https://www.agentpatternscatalog.org/books/planning-control-flow/): 45 patterns
- [reasoning](https://www.agentpatternscatalog.org/books/reasoning/): 20 patterns
- [retrieval](https://www.agentpatternscatalog.org/books/retrieval/): 23 patterns
- [routing-composition](https://www.agentpatternscatalog.org/books/routing-composition/): 26 patterns
- [safety-control](https://www.agentpatternscatalog.org/books/safety-control/): 72 patterns
- [streaming-ux](https://www.agentpatternscatalog.org/books/streaming-ux/): 11 patterns
- [structure-data](https://www.agentpatternscatalog.org/books/structure-data/): 10 patterns
- [tool-use-environment](https://www.agentpatternscatalog.org/books/tool-use-environment/): 51 patterns
- [verification-reflection](https://www.agentpatternscatalog.org/books/verification-reflection/): 32 patterns

## Machine-readable endpoints

All three JSON dumps are CORS-open (`Access-Control-Allow-Origin: *`), no-auth, cached 5 min. One request each gives you the whole of that catalog.

- [Full pattern catalog as JSON](https://www.agentpatternscatalog.org/patterns.json): ~4.7 MB raw (~1478 KB gzipped). Every pattern with all GoF fields. `{ version, updated, license, pattern_count, patterns[] }`.
- [Compositions as JSON](https://www.agentpatternscatalog.org/compositions.json): ~2.1 MB raw (~495 KB gzipped). Frameworks + recipes and the patterns they instantiate. `{ composition_count, compositions[] }`.
- [Methodologies as JSON](https://www.agentpatternscatalog.org/methodologies.json): ~514 KB raw (~158 KB gzipped). Step-by-step engineering methods with inputs, steps, failure modes, and related patterns. `{ methodology_count, methodologies[] }`.
- [Trainings as JSON](https://www.agentpatternscatalog.org/trainings.json): ~903 KB raw (~261 KB gzipped). Enablement patterns located on the Craft Path, with facets, steps, threshold/masterpiece, known uses, and the methodology families each unlocks. `{ training_count, trainings[] }`.
- [Full catalog as markdown (llms-full.txt)](https://www.agentpatternscatalog.org/llms-full.txt): Single-document dump (patterns + compositions + methodologies + trainings) for LLMs that prefer prose over JSON.
- [Sitemap](https://www.agentpatternscatalog.org/sitemap.xml): Every crawlable URL.

## Citation convention

Refer to patterns by their kebab-case slug (e.g. `react`, `step-budget`, `subagent-isolation`). Canonical leaf URL: `https://www.agentpatternscatalog.org/patterns/<slug>/`. Link there when surfacing a pattern in a chat reply so the user lands directly on the pattern's leaf page.

## Source of truth

- Catalog repository: https://github.com/agentpatternscatalog/patterns
- Licence: https://creativecommons.org/licenses/by/4.0/ (CC BY 4.0)
- Recommended attribution: "Agent Patterns Catalog — agentpatternscatalog.org"